Bug 813819

This code in libvirt.c is bogus:

    if (interval <= 0) {
        virLibConnError(VIR_ERR_INVALID_ARG,
                        _("negative or zero interval make no sense"));
        goto error;
    }

AFAICT from GIT history this check has been there since day 1, so I'm sceptical that this API worked as you described in 6.2

Also note that if you don't register any main loop impl with libvirt, then it should automatically skip usage of keep alive. The problem only arises if you register a main loop impl, but then don't run it.

Sorry, I expressed myself incorrectly. The connection dying because the
keep-alive message are not send is new thing in 6.3. Calling
"virConnectSetKeepAlive" is how I tried to fix that.

Does your application register a main loop implementation with libvirt ?  If it does not, then keep alive should not even have activated.

I think I nailed down the issue. This hasn't changed between 6.2 and 6.3, sorry. It was my fault in mixing threads and forking. I started the event loop before double-fork when starting the daemon and it caused that keep-alive messages get lost.

Nevertheless the documentation is still wrong and should be changed. But that's not 6.3 stuff, feel free to close this bug.

> Nevertheless the documentation is still wrong and should be changed. But that's
> not 6.3 stuff, feel free to close this bug.

Actually the code is wrong, so we should still fix this problem

Fixed upstream:
commit 6446a9e20cc65561ce6061742baf35a3a63d5ba1
Author: Peter Krempa <pkrempa>
Date:   Tue Apr 24 16:38:41 2012 +0200

    keepalive: Add ability to disable keepalive messages

Hi, 

I will try to verify this bug when new build is out so could you provide the method to verfy this bug ? 
thanks very much

Wenlong

Test with libvirt-0.9.10-11.el6 and libvirt-0.9.10-21.el6:

set interval=-1 in libvirtd.conf
keepalive_interval = -1
keepalive_count = 0

run the event-test.py
# python /usr/share/doc/libvirt-python-0.9.10/events-python/event-test.py 

get the following msg in libvirtd.log:

2012-06-20 06:23:36.547+0000: 21614: debug : virKeepAliveNew:244 : client=0x11c5db0, interval=-1, count=0
2012-06-20 06:23:36.547+0000: 21614: debug : virKeepAliveNew:277 : RPC_KEEPALIVE_NEW: ka=0x11c5b70 client=0x11c5db0 refs=2
2012-06-20 06:23:36.548+0000: 21614: debug : virKeepAliveCheckMessage:408 : ka=0x11c5b70, client=0x11c5db0, msg=0x7f7e4e546010
2012-06-20 06:23:36.550+0000: 21614: debug : virKeepAliveCheckMessage:408 : ka=0x11c5b70, client=0x11c5db0, msg=0x7f7e4e505010
2012-06-20 06:23:36.550+0000: 21616: debug : virKeepAliveStart:346 : Keepalive messages disabled by configuration

Seems that the interval<0 has taken affect but cann't get the "negative or zero interval make no sense".

I also notice that the code was removed in the upstream and " * Note: This API function controls only keepalive messages sent by the client."  for virConnectSetKeepAlive. 
Not sure whether we can test it with event-test.py as the client. Can you help to confirm that how can we trigger this error functionally? then we can reproduce and verify it once it's ON_QA. Thanks.

Hi, rnovacek

Can you help to provide some steps to reproduce this bug ? thanks.

To get this "negative or zero interval make no sense" error in non fixed version of libvirt run this python script:

import libvirt
v = libvirt.open("")
v.setKeepAlive(-1, 1)

To test setting positive keep-alive interval you need event loop in client, you can start it with this script:

eventLoopThread = None

def virEventLoopNativeRun():
    import libvirt
    while True:
        libvirt.virEventRunDefaultImpl()

def virEventLoopNativeStart():
    global eventLoopThread
    libvirt.virEventRegisterDefaultImpl()
    eventLoopThread = threading.Thread(target=virEventLoopNativeRun, name="libvirtEventLoop")
    eventLoopThread.setDaemon(True)
    eventLoopThread.start()

virEventLoopNativeStart()

To test if the interval is correct you might try to stop the event loop then wait longer than the keep-alive interval and check if the connection breaks. But I'm not sure it will work, this is only my assumption.

While developing that functionality I used a trivial libvirt client written in C that enabled keepalives and manually called the event loop a few times and then disabled them and run the event loop again. Unfortunately I can't find it anymore.

To test this you need to disable server side keepalives, as they could interfere with client keepalives.

Verify this bug with : 

libvirt-0.10.0-0rc0.el6.x86_64

terminal1: 
# python /usr/share/doc/libvirt-python-*/events-python/event-test.py
Using uri:qemu:///system

terminal2: 

# ./t.py 
libvir: Remote Driver error : internal error the caller doesn't support keepalive protocol; perhaps it's missing event loop implementation
Traceback (most recent call last):
  File "./t.py", line 5, in <module>
    v.setKeepAlive(-1, 1)
  File "/usr/lib64/python2.6/site-packages/libvirt.py", line 3150, in setKeepAlive
    if ret == -1: raise libvirtError ('virConnectSetKeepAlive() failed', conn=self)
libvirt.libvirtError: internal error the caller doesn't support keepalive protocol; perhaps it's missing event loop implementation


#cat t.py 
#!/usr/bin/python

import libvirt
v = libvirt.open("")
v.setKeepAlive(-1, 1)

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

http://rhn.redhat.com/errata/RHSA-2013-0276.html